[NEW] AWS Certified AI Practitioner [2026]

Concepts of supervised, unsupervised, and reinforcement learning.

Conceptual understanding of classification, regression, and clustering algorithms.

The ML lifecycle: data collection, preparation, training, evaluation, deployment, and monitoring.

Practical use cases: forecasting, recommendation, anomaly detection, computer vision, and NLP.

Fundamentals of Generative AI (24%)

Large language model (LLM) architectures and training methods.

Use cases for text generation, image synthesis, and code assistance.

Prompt engineering basics and advanced techniques.

Model fine-tuning and evaluation metrics for generative outputs.

Applications of Foundation Models (28%)

Selecting the right AWS services (e.g., Amazon Bedrock, Amazon SageMaker).

Evaluating performance and cost for foundation model inference.

Implementing prompt techniques for various modalities.

Assessing scalability and latency for production workloads.

Guidelines for Responsible AI (14%)

Inclusivity and the importance of diverse training data.

Transparency, explainability, and interpretability in AI models.

Safety, robustness, and human oversight mechanisms.

Legal, ethical, and compliance implications.

Security, Compliance, and Governance for AI Solutions (14%)

Implementing data lineage tracking and Model Cards.

Applying privacy-enhancing techniques for data at rest and in transit.

Aligning with regulatory frameworks using AWS Config, Audit Manager, and Artifact.

Establishing strict governance policies and monitoring systems.

B. Amazon EC2

C. Amazon Bedrock

D. AWS Lambda

E. Amazon Comprehend

F. Amazon Textract

Correct Answer: C

Overall Explanation: Amazon Bedrock is a fully managed service that offers a choice of high-performing foundation models from leading AI companies via a single API, making it the perfect choice for building generative AI applications without managing infrastructure.

Option A is incorrect: While Amazon SageMaker can host foundation models (via SageMaker JumpStart), it requires you to provision, manage, and scale the underlying infrastructure (endpoints), which goes against the requirement of not managing servers.

Option B is incorrect: Amazon EC2 provides raw virtual servers. You would have to install, configure, and maintain the models and infrastructure entirely from scratch.

Option C is correct: Amazon Bedrock is explicitly designed to provide serverless access to foundation models via API.

Option D is incorrect: AWS Lambda is a serverless compute service. While it can trigger API calls, it is not an AI service that hosts foundation models.

Option E is incorrect: Amazon Comprehend is an NLP service used for finding insights and relationships in text (like sentiment analysis), not a generative AI service for document summarization using foundation models.

Option F is incorrect: Amazon Textract is an ML service used to extract text, handwriting, and data from scanned documents, not a generative AI summarization tool.

Question 2: You are evaluating an ML model deployed in production to ensure it aligns with responsible AI guidelines. You notice the model's predictions disproportionately favor a specific demographic. Which concept of responsible AI does this issue primarily violate, and what should be used to document the model's intended use and characteristics?

B. Privacy; implement AWS KMS encryption.

C. Fairness; implement Model Cards.

D. Robustness; implement Multi-AZ deployments.

E. Explainability; implement AWS Config rules.

F. Security; implement IAM resource policies.

Correct Answer: C

Overall Explanation: Responsible AI dictates that models should treat all users fairly. When a model favors one demographic over another, it exhibits bias, which violates the principle of fairness. AWS recommends using Model Cards (like Amazon SageMaker Model Cards) to document a model's intended use, risk ratings, and training characteristics.

Option A is incorrect: Transparency is important, but favoring a demographic is a bias issue, not strictly a logging/transparency issue. CloudWatch logs system metrics, not model intent.

Option B is incorrect: Privacy and KMS encryption protect data from unauthorized access, but they do not solve demographic bias in predictions.

Option C is correct: The scenario describes a fairness and bias issue. Model Cards are the AWS standard for documenting model governance, limitations, and fairness metrics.

Option D is incorrect: Robustness in AI often refers to handling edge cases or adversarial inputs. Multi-AZ is for high availability of infrastructure, not model fairness.

Option E is incorrect: Explainability is the ability to understand how a model makes a decision. While helpful, the core violation here is fairness. AWS Config tracks resource configuration changes, not AI explainability.

Option F is incorrect: Security and IAM protect against unauthorized access, which is unrelated to demographic bias in the algorithm's outputs.

Question 3: According to AWS security, compliance, and governance best practices, how should an organization ensure that only authorized applications can invoke generative AI models while keeping the prompt data strictly private?

B. Send prompt data over the public internet to third-party endpoints.

C. Use IAM roles with least privilege and ensure data is encrypted at rest and in transit.

D. Disable AWS CloudTrail to prevent prompt data from being logged or audited.

E. Store all prompt data in an unencrypted Amazon S3 bucket for faster retrieval.

F. Hardcode long-term IAM user credentials directly into the application's source code.

Correct Answer: C

Overall Explanation: Security in AWS AI workloads relies on the principle of least privilege using IAM, combined with strong encryption protocols for data both at rest and in transit.

Option A is incorrect: Providing broad AdministratorAccess violates the principle of least privilege and introduces massive security risks.

Option B is incorrect: Sending private prompt data over the public internet without proper VPC endpoints (like AWS PrivateLink) compromises data privacy.

Option C is correct: IAM roles grant specific, temporary permissions (least privilege), and encryption protects the privacy of the prompts.

Option D is incorrect: AWS CloudTrail should be enabled for governance and auditing purposes. Disabling it reduces visibility and violates compliance frameworks.

Option E is incorrect: Storing sensitive data in unencrypted S3 buckets directly violates privacy and compliance guidelines.

Option F is incorrect: Hardcoding credentials is a severe security anti-pattern. IAM roles and temporary credentials should always be used for applications.

You can retake the exams as many times as you want

This is a huge original question bank

You get support from me if you have questions

Each question has a detailed explanation

Mobile-compatible with the Udemy app